A global test can be an attractive alternative to multiple hypothesis tests in a trial with more than one important measured outcome. Such a test typically uses a non-parametric combination of several outcomes in one combined measure. However, adequately controlling type I error could involve non-trivial additional computations in the presence of the non-parametric Behrens-Fisher problem. In addition, determining the sample size for a trial using a global test may also present challenges. Thus, it is a question whether the advantages of such a test outweigh the difficulties in its implementation.

Q1: Have you ever used or recommended a global test in a regulatory trial? Q2: Can you describe any situations where a global test might have been useful, but was not used? Q3: Are you aware of any research on the use of global tests with highly correlated component domains?